Tiiny AI, a U.S.-based startup focusing on personal AI hardware, has introduced a compact mini PC designed to run large language models locally. The device, called Pocket Lab, is built around a 12-core ARM processor and an NPU rated at up to 160 trillion operations per second (TOPS), supported by 80 GB of LPDDR5X memory.
The company says the system’s performance enables users to run models with as many as 120 billion parameters. In its announcement, Tiiny AI stated that Pocket Lab has been recognised by Guinness World Records as the “world’s smallest mini PC” in what it described as the “100B LLM Local category,” a designation that appears highly specialised but underscores the device’s compact form. Measuring 14.2 × 8 × 2.5 cm and weighing 300 grams, the system is designed to fit in a pocket.
See also: Valve Signals Steam Machine Pricing Will Align With Gaming PCs, Not Consoles
Tiiny AI says the Pocket Lab supports “one-click installation” for a range of open-source models, including Llama, Qwen, DeepSeek, Mistral, Phi and OpenAI GPT-OSS, with over-the-air updates planned. The hardware includes a 1 TB SSD and operates within a 65-watt power envelope, with the company emphasising fully offline capability.
While Tiiny AI claims the mini PC can handle 120-billion-parameter models, the quantization levels were not disclosed. The company said the device “packs quite a punch,” but real-world performance remains untested, and running such large models on 80 GB of memory may prove challenging even with aggressive quantization.
See also: Aoostar Brings AMD Ryzen AI Max+ 395 mini-PC to Worldwide Audience
Pricing has not yet been announced, though the device is expected to be expensive. For users without demanding local AI workloads, conventional mini PCs such as the Minisforum AI X1 Pro may remain more practical options.
Source: TechPowerUp